What is human in the loop?

A design where a person must approve an agent's decision before it takes effect.

How does it work?

The agent proposes. A person reviews and clicks approve or reject. The agent acts only on approval. Vendors describe this as the safe default, and most agentic products in revenue work ship with it on.

Why does it matter for agents?

The placebo essay’s argument is that the checkpoint often adds delay without judgment. The reviewer sees only what the agent shows and approves nearly everything unchanged. The cost of that delay is the autonomy tax, and the calculator on this site puts a number on it.

What replaces it?

Boundaries instead of approvals. People decide what the agent can and cannot do and write those limits down as stop conditions. They handle the real exceptions and adjust the limits from results in aggregate. The human stays in charge of the rules and leaves the loop.

Which essays cover it?

• Human-in-the-Loop Is a Temporary Placebo
• Why Your AI Agents Are Failing: The Control Trap
• Stop Conditions Are the Product

Which terms are related?

• autonomy tax
• stop condition

@misc{couch2026whatishumanintheloop,
  author = {Couch, Chris},
  title = {What is human in the loop?},
  year = {2026},
  howpublished = {Ends at Cash},
  url = {https://chriscouch.ai/what-is/human-in-the-loop/}
}